Content
75%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A well-structured four-phase workflow with a concrete API endpoint, exact p75 thresholds, a comparison table template, explicit stop conditions, and a fallback path. The gaps are minor: the intro duplicates the description's premise, no example CrUX request payload is given, and partial-data cases plus content splitting into references go unaddressed.
Suggestions
Trim or fold the opening paragraph into a single goal line — it re-explains the lab-vs-field premise the description already states.
Add a copy-paste example CrUX records:queryRecord request (curl with JSON body for one origin/URL) so Phase 0 is fully executable without recalling the API shape.
Address partial-data checkpoints in the workflow, e.g., what to do when origin-level CrUX data exists but URL-level data is missing.
Consider moving the CrUX API query details or the Phase 2 divergence-cause catalog into a one-level-deep reference file to keep SKILL.md a lean overview.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient — the Phase 2 divergence causes are genuinely non-obvious domain knowledge that earns its tokens — but the 3-sentence opening paragraph ('Lighthouse measures one page load, once...') restates the description's premise and could be trimmed, and parenthetical justifications ('matching Google's own "Good" threshold methodology') are mildly padded. | 4 / 5 |
Actionability | Concrete guidance includes the actual CrUX endpoint URL, the API-key/PSI-fallback path, named RUM sources to check (GA4 Web Vitals event, Cloudflare, Sentry, Datadog), exact p75 thresholds (<2.5s, <200ms, <0.1), and a comparison table template. The minor gap is that no example API request payload (curl or JSON body) is shown, so the Phase 0 data pull is not fully copy-paste ready. | 4 / 5 |
Workflow Clarity | Phases 0–3 are clearly sequenced with real checkpoints — 'that absence IS the Phase 3 finding. Stop here rather than fabricating', the PSI fallback when no API key exists, and 'If field is *better* than lab, note it and move on'. Not a destructive/batch operation so no validation cap applies; remaining gaps are minor (e.g., handling origin-level CrUX data when URL-level data is missing). | 4 / 5 |
Progressive Disclosure | No bundle files exist and none are strictly needed; sections are well-organized and external skill pointers ('load-test-bootstrap', 'third-party-script-audit') are clearly signaled. At ~82 lines it exceeds the under-50-line simple-skill exception for a top score, and material like the CrUX API query details or the divergence-cause list could plausibly live in a one-level-deep reference file. | 4 / 5 |
Total | 16 / 20 Passed |